data: Cu10 CUTLASS headers from corex-samples

cat_files/default_mma_core.h

@@ -0,0 +1,114 @@
+/***************************************************************************************************
+ * Copyright (c) 2017-2021, NVIDIA CORPORATION.  All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without modification, are permitted
+ * provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright notice, this list of
+ *       conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright notice, this list of
+ *       conditions and the following disclaimer in the documentation and/or other materials
+ *       provided with the distribution.
+ *     * Neither the name of the NVIDIA CORPORATION nor the names of its contributors may be used
+ *       to endorse or promote products derived from this software without specific prior written
+ *       permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR
+ * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
+ * FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL NVIDIA CORPORATION BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
+ * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
+ * STRICT LIABILITY, OR TOR (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ **************************************************************************************************/
+
+/***************************************************************************************************
+* Copyright (c) 2021 Iluvatar CoreX. All rights reserved.
+* Copyright Declaration: This software, including all of its code and documentation,
+* except for the third-party software it contains, is a copyrighted work of Shanghai Iluvatar CoreX
+* Semiconductor Co., Ltd. and its affiliates ("Iluvatar CoreX") in accordance with the PRC Copyright
+* Law and relevant international treaties, and all rights contained therein are enjoyed by Iluvatar
+* CoreX. No user of this software shall have any right, ownership or interest in this software and
+* any use of this software shall be in compliance with the terms and conditions of the End User
+* License Agreement.
+ **************************************************************************************************/
+
+/*! \file
+    \brief Defines basic properties needed by CTA-level GEMMs assuming expectations about data
+      layout of the global memory fragments, data types, and internal tile sizes.
+
+      Partial specializations for threadblock::Mma operations targeting TensorOp instructions.
+*/
+
+#pragma once
+
+#include "cutlass/cutlass.h"
+#include "cutlass/array.h"
+
+#include "cutlass/numeric_types.h"
+#include "cutlass/matrix_shape.h"
+
+#include "cutlass/gemm/warp/mma.h"
+#include "cutlass/gemm/threadblock/mma_pipelined.h"
+#include "cutlass/gemm/threadblock/mma_singlestage.h"
+#include "cutlass/gemm/threadblock/mma_preload.h"
+#include "cutlass/arch/cache_operation.h"
+
+/////////////////////////////////////////////////////////////////////////////////////////////////
+
+namespace cutlass {
+namespace gemm {
+namespace threadblock {
+
+/////////////////////////////////////////////////////////////////////////////////////////////////
+
+/// Template defininng default matrix multiply operators inferred from threadblock tile size,
+/// global memory data layout, and target math instruction.
+template <
+    /// Shape of threadblock-scoped matrix multiply operator
+    typename Shape,
+    /// Shape of warp-level matrix multiply operator
+    typename WarpShape,
+    /// Shape of one matrix production operation (concept: GemmShape)
+    typename InstructionShape,
+    /// Element data type of A operand
+    typename ElementA,
+    /// Layout of operand A
+    typename LayoutA,
+    /// Element data type of B operand
+    typename ElementB,
+    /// Layout of operand B
+    typename LayoutB,
+    /// Data type of accumulator
+    typename ElementC,
+    /// Layout of accumulator
+    typename LayoutC,
+    /// Indicates type of math operator (arch::OpClassSimt or arch::OpClassTensorOp)
+    typename OperatorClass,
+    /// Number of stages
+    int Stages = 2,
+    /// Operation performed by MMA
+    typename Operator = cutlass::arch::OpMultiplyAdd,
+    /// Store the accumulators in row major or column major.  Row major is used
+    /// when output layout is interleaved.
+    bool AccumulatorsInRowMajor = false,
+    /// Cache operation of operand A
+    cutlass::arch::CacheOperation::Kind CacheOpA =
+        cutlass::arch::CacheOperation::Global,
+    /// Cache operation of operand B
+    cutlass::arch::CacheOperation::Kind CacheOpB =
+        cutlass::arch::CacheOperation::Global,
+    /// per-element transformation for elements of A
+    ComplexTransform TransformA = ComplexTransform::kNone,
+    /// per-element transformation for elements of B
+    ComplexTransform TransformB = ComplexTransform::kNone,
+    bool IsComplex = false // (is_complex<ElementA>::value || is_complex<ElementB>::value)
+>
+struct DefaultMmaCore;
+
+/////////////////////////////////////////////////////////////////////////////////////////////////
+
+} // namespace threadblock
+} // namespace gemm
+} // namespace cutlass